A Novel Method for Sleep Score Estimation Using Wearable Sensors with a Deep Sequential Neural Network

摘要

The use of sleep score as a measure of fitness and wellness is getting popular in Smart Health as it provides an objective assessment of sleep quality. However, reliable estimation of sleep scores from wearable sensor data only is challenging. In this study, we investigated the estimation of sleep score using only features available from single-channel ECG or single-channel EEG data. We used partial correlation and conditional permutation importance for feature selection; then compared extreme gradient boosting, artificial neural network, and sequential neural network for developing a regression model for sleep score estimation. TabNet- an attention-based deep sequential learning model achieved the best performance of RMSE = 5.47 and R-squared value of 0.59 in the test set for sleep score estimation using only spectral features of single-channel EEG. The results pave the way for reliable and interpretable sleep score estimation using a wearable device.